Claude Code Agent View: Parallel Agents Are Here
Leon van Zyl ·2026-05-14 ·2 min read
Summary written by us from the video's transcript. The video, and everything in it, is Leon van Zyl's work.
Learn how Claude Code’s new Agent View lets you run, monitor, and manage multiple background agents from a single interface, including permissions handling, scheduling commands, and integrating with Git worktrees.
Takeaways
- Agent View centralizes all Claude Code agents, letting you launch, monitor, and switch between them without multiple terminals.
- Permission modes can be toggled on‑the‑fly, enabling fully automated background work or interactive approval as needed.
- Built‑in commands like **loop** and **goal** let you schedule recurring improvements or run long‑running tasks unattended.
- You can stop all agents with Ctrl + X and later resume them, preserving work state across sessions.
- Agent View works across multiple projects and integrates smoothly with Git worktrees for rapid design iteration.
What is Agent View and why it matters
Agent View adds a dispatcher window that shows all active Claude Code sessions at a glance, eliminating the need for separate terminal windows when running parallel agents. It works on any OS after updating your Cloud Code instance to the latest version.
Starting Agent View and launching background agents
Run `cloud agents` (or `cloud --dangerously-skip-permissions --bg`) to open the dispatcher. From there you can give an instruction—e.g., “update the logo with a suitable emoji”—and a new Claude Code station is created in the background.
You can also start a background agent directly from any terminal with `claud bg "Update the readme file"`; the command creates a hidden station that appears later in the dispatcher.
Navigating sessions and managing permissions
Use the mouse or arrow keys to select a station, press Enter to enter it, and the back arrow (or Backspace) to return to the dispatcher. Sessions can be collapsed/expanded, and completed jobs move to a separate tab.
Permission modes control whether Claude asks before making changes. Switch modes with Shift + Tab to “accept all edits” or run agents in “dangerous” mode by adding `--dangerously-skip-permissions` when launching.
Running advanced commands: loop, goal, and skills
The **loop** command creates a cron‑like job that runs every X minutes, identifies one meaningful improvement, makes the change, commits it, and repeats automatically.
The **goal** command handles complex, long‑running tasks (e.g., adding an issue‑tracking feature) until completion, similar to the older Ralph command.
You can invoke custom skills—such as a security scanner—directly from Agent View, optionally setting auto‑accept mode so the skill runs without repeated prompts.
Stopping, resuming, and viewing across projects
Press Ctrl + X in the dispatcher to stop all running sessions; they move to the Completed tab with a “stopped” status. Stopped sessions can be resumed by selecting them and choosing “resume.”
Agent View aggregates sessions from multiple Claude Code projects, so you can monitor work in different folders (e.g., a Get‑Images app and an AI‑To‑Do app) without opening each project separately.
Using Agent View with Git worktrees for design exploration
Create a base project, then launch `cloud --dangerously-skip-permissions --bg "Please scaffold a very simple to‑do app using React and Vite…"` to start scaffolding in a background agent.
Copy prompts that describe different designs and include instructions to create isolated worktrees. After each worktree finishes, add a command to run the dev server on a unique port, then open the URLs to compare designs side by side.